Building an AI data centre with Oracle

How do you build an AI data centre? Last week, I had the absolute pleasure of hosting a fireside chat with Dan Madrigal. Here's what I learned.

AI data centres

AI data centres have constantly made the headlines. They are extremely expensive; they are crucial for AI training... and very few know how to build them.

So I was delighted at the chance to chat on stage at W.Media CDC with Dan, VP for Cloud AI Sourcing at Oracle, given his direct experience building some of the largest AI data centres on the planet.

But what's the big deal with Oracle and AI? I explain here.

Here are three things I walked away with.

Doing 4D math

Despite having built data centres for some decades now, we are all new to AI data centres. In our chat, Dan talked about grappling with new considerations such as:

• No track record for systems on maiden deployments.
• How will extreme strain impact equipment lifecycles?
• Depreciation of such heavily loaded systems.

So yes, new "unproven" systems designed just for AI, exceedingly heavy use never seen before, and still need to calculate a sensible depreciation.

"You have to think about it multi-dimensionally, and different life cycle times to actually deliver, operate and maintain these assets [well] - and eventually replace some of those components."

I just call it doing 4D math.

An educated 'bet'

But just because systems or deployments are new doesn't mean the team gets a free pass when things go wrong. Which means the onus is on them to back up any decision they make.

Some considerations:

• Make an educated guess that's based on science.
• Look at the track record of contractor, developer.
• Make the best decision based on customer needs.

Then there's the financial aspects:

• The underwriting.
• The returns on the asset.
• The margins as a company.

"You have to keep it multi-dimensional, and you have to continue iterating. And the speed of change is so fast that it's so hard to figure out."

No wonder Dan joked about waking up in cold sweat at night, frantic about having forgotten something.

Advice for DC professionals

But perhaps the part of the conversation that struck me the most was Dan sharing how he "stumbled" into the data centre industry by accident.

Turns out it's true for many in the room.

His advice for budding data centre professionals?

"Lifelong learning is important. You grow by learning. And if you stop learning, you stop growing. And there's so much to learn. Be curious and harness that curiosity."